Penentuan Skala Prioritas Penagihan Rekening Listrik Untuk Mencapai Zero Debt Menggunakan Random Forest Dan Decision Tree

Wisuda, Heri Priyo (2024) Penentuan Skala Prioritas Penagihan Rekening Listrik Untuk Mencapai Zero Debt Menggunakan Random Forest Dan Decision Tree. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6032212129-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 April 2026. Download (1MB) | Request a copy

Abstract

PLN merupakan perusahaan BUMN yang bertugas melistriki hingga ke pelosok tanah air. PLN secara umum melayani 2 jenis pelanggan yaitu pelanggan prabayar dan pelanggan pasca bayar. PLN Unit XYZ pada tahun 2022 mengelola 2.100.218 pelanggan dengan komposisi 1.218.126 atau 57,99% pelanggan prabayar dan 882.092 atau 42,01% pelanggan paska bayar. Pelanggan paska bayar yaitu pelanggan yang memakai energi listrik terlebih dahulu dan melakukan pembayaran di bulan berikutnya. Pelanggan yang membayar tagihan rekening listrik diatas tanggal 20 atau pelanggan menunggak sebesar 193 ribu atau sebesar 22%. Hal tersebut menghambat percepatan cash in dari pendapatan penjualan energi listrik. Penagihan pelanggan menunggak menggunakan tenaga alih daya/billman dengan diberikan work order penagihan. Proses penagihan pelanggan menunggak masih bersifat sporadic, tanpa peng-klasifikasian atau belum menggunakan skala prioritas sehingga tidak optimal. Dengan frame work CRISP – DM (Cross Standard Industry for Data Mining), fitur dari data pelanggan dapat digunakan untuk memprediksi pelanggan menunggak atau bukan menggunakan machine learning random forest dan decision tree. Hasil pemodelan random forest dan decision tree memiliki kinerja yang sama dengan ditunjukkan nilai AUC, Precision dan Recall sama besar. Pemodelan baik decision tree dan random forest kecenderungan memprediksi kelas tidak menunggak atau kelas mayoritas. Penggunaan SMOTE untuk over sampling menaikkan nilai precision dan Recall, sehingga menaikan jumlah True Positif kelas menunggak. Hasil prediksi random forest dan decision tree meningkatkan efisiensi jumlah lembar tunggakan yang harus ditagihkan ke pelanggan serta mempercepat cash in perusahaan untuk mencapai zero debt dan menciptakan budaya baru yaitu membayar listrik tepat waktu.
========================================================================================================================
PLN is a state-owned company responsible for providing electricity throughout the country. In general, PLN serves two types of customers: prepaid and postpaid customers. In the year 2022, PLN Unit XYZ managed 2,100,218 customers, with a composition of 1,218,126 or 57.99% prepaid customers and 882,092 or 42.01% postpaid customers. Postpaid customers are those who consume electricity first and make payments in the following month. Customers who pay their electricity bills after the 20th of the month or have arrears amount to 193,000 or 22%. This hampers the acceleration of cash flow from electricity sales revenue. The collection of overdue customers is done using outsourced manpower/billman with the issuance of work orders. The process of collecting overdue customers is still sporadic, without classification or prioritization scale, thus not being optimal. With the CRISP-DM (Cross Industry Standard Process for Data Mining) framework, customer data features can be used to predict whether customers will default or not using machine learning random forest and decision tree algorithms. The modeling results of both random forest and decision tree show the same performance with equal values of AUC, Precision, and Recall. Both decision tree and random forest modeling tend to predict the non-default class or the majority class. The use of SMOTE for oversampling increases the precision and Recall values, thus increasing the number of True Positives for the default class. The prediction results of random forest and decision tree improve the efficiency of the number of overdue bills that need to be billed to customers and accelerate the company's cash flow to achieve zero debt and create a new culture of paying electricity bills on time.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Pelanggan menunggak, crisp-dm, machine learning, random forest, decision tree, SMOTE, overdue customer
Subjects:	H Social Sciences > HA Statistics > HA31.3 Regression. Correlation
Divisions:	Interdisciplinary School of Management and Technology (SIMT) > 61101-Master of Technology Management (MMT)
Depositing User:	HERI PRIYO WISUDA
Date Deposited:	23 Feb 2024 07:18
Last Modified:	23 Feb 2024 07:18
URI:	http://repository.its.ac.id/id/eprint/107692

Actions (login required)

View Item